Op — Deep research method

Version 1.0.0
sha256: 332c9dac3bfcfb32c6529225e100c48ce8e992b175a396d80edbbe820177e084

---
name: op-deep-research-method
description: Use for any research task with a live-world component: market, ecosystem, competitor, spec, or "what changed" questions. Produces primary-source evidence with dated fetches and explicit uncertainty.
---
# Deep research method

1. Frame the decision the research serves in one sentence; list the 5–8 claims that would change it.
2. Primary sources only for load-bearing claims: official docs, specs, RFCs, official repos, filings. Blogs and news are corroboration, never the sole source.
3. Fetch every source you cite. Record the URL, the fetch date, and a verbatim quote under 15 words or a precise paraphrase. Mark anything unfetchable UNVERIFIED — never fill the gap from memory.
4. Parallelize independent threads across subagents; each returns a ledger, not prose.
5. Check versions and dates: specs, CLIs and docs move monthly — look for a newer version than the one you remember.
6. Weigh the evidence by source quality and recency; state confidence High/Med/Low per claim.
7. Write the synthesis in prose, evidence first, then the opinion, then what remains uncertain.
8. Second pass: re-fetch the top 5 load-bearing sources and try to falsify the conclusion.